The DOM samples were collected from the Great Dismal Swamp, in Virginia, USA, and from the Suwannee River, in Florida, USA. The Great Dismal Swamp was sampled in Suffolk (36.7∘ S, 76.4∘ W) in 2014 (Sun et al., 2014) and from Jericho Ditch (36.7∘ S, 76.4∘ W) in 2016 (Lin et al., 2017). Suwannee River water was collected in 2017 close to the collection site of the International Humic Standard Society (IHSS) (30.5∘ N, 82.5∘ W). The Great Dismal Swamp and the Suwannee River samples were filtered through a pre-cleaned 0.2 µm capsule filter (Polycap TC, Whatman) on-site following collection and kept refrigerated until use. The samples were never frozen to avoid potential freeze–thaw cycle artefacts. Experimental solutions with concentrations of approximately 20 mg C L-1 were prepared upon dilution of the field-collected waters in nanopure water. The Suwannee River Fulvic Acid Standard II (2S101F) was purchased from the IHSS and dissolved in nanopure water with concentrations of approximately 20 mg C L-1. The inorganic ions present in Great Dismal Swamp water have been previously reported (Johannesson et al., 2004), and anionic ion chromatography confirmed similar concentrations. Indeed, DS2014 had concentrations of F-, Cl-, SO42- and NO3- for a total of 30 µM, which we can approximate to 60 µM of ions assuming charge balance. These ion concentrations suggest that the ratio of organic to inorganic ions is typically less than 10 % by molar concentration.

In addition, the DOM sampled for this study was bulk water-collected and therefore represents an alternative sampling procedure to access organic carbon compared to impacting particular matter on filters. Indeed, DOM was not extracted from a filter or from an impinger but rather directly collected in the field. Ultimately, the photochemical mechanisms identified in this study are expected to be similar between DOM and organic aerosols, thereby bridging the fields of aquatic and atmospheric chemistry of organic matter.

Photolysis experiments were conducted in a commercial photoreactor (Rayonet, Southern New England Ultraviolet Co.) equipped with a motorized turntable and 6×300 nm light bulbs (UVB – 3000 Å from Southern New England Ultraviolet Co.). These UVB bulbs have a relative light intensity as a function of wavelength shown in Fig. S2. A total of 9 mL of a 20 mg C L-1 DOM or IHSS isolate solution was pipetted into cork-capped 10 mL borosilicate test tubes (Pyrex, 15 mm × 85 mm, disposable) and irradiated for up to 25 h. During this period, the temperature inside the photoreactor was held at 30–32 ∘C, and the reactor was turned on prior to the beginning of the experiment to ensure a constant temperate exposure. At each time point, a test tube was removed from the photoreactor, and the solution was directly used for either CCN (9–18 mL), INP (9–18 mL), total organic carbon (TOC; 7 mL), ion chromatography (1 mL) or conductivity (0.1 mL) measurements.

A slightly different experimental setup was used for the CO and CO2 measurements. In order to avoid CO2 partitioning to the gas phase, the borosilicate test tubes were prepared headspace-free using rubber septa. At each time point, a test tube was withdrawn from the photoreactor and 6 mL of the experimental solution was transferred to a N2-flushed 20 mL serum vial containing 100 µL HCl 1 N. The serum vial was briefly shaken and stored no longer than a week in the fridge until headspace analysis on the gas chromatograph equipped with a flame ionization detector (GC-FID).

Control experiments were conducted to unambiguously attribute the change in CCN and INPs to photochemistry. Importantly, the photomineralization mechanism could have led to the accumulation of carbonate ions in solution and thus increase the CCN ability of the material with UVB exposure time. As a control, inorganic carbon measurements by the Shimadzu TOC analyser showed no increasing IC fraction with irradiation. In addition, control experiments where the DOM solutions were purged with argon showed no difference in CCN activity, indicating no contribution to accumulating dissolved carbonate in solution.

To address the issue of the potential impact of a 30 ∘C heat exposure on CCN and INPs during the photochemical experiment, foil-covered test tubes were placed inside the photoreactor alongside non-covered test tubes. Indeed, the non-irradiated sample that would have experienced the same temperature elevation from room temperature to 30 ∘C did not show any change in hygroscopicity, INP activity, organic carbon or production of photochemical adducts.

Control experiments involving Ar-purged solutions further emphasize that dissolved O2 is necessary for photomineralization, consistent with previous reports (Schmitt-Kopplin et al., 1998). In an atmospheric context, O2 will be readily available to the aqueous aerosol phase, and we expect in situ aqueous photomineralization of organic aerosols. Furthermore, as confirmed by dark control experiments, exposure to light was necessary to drive the chemical, cloud droplet formation and ice nucleation changes observed.

The light intensity inside the photoreactor was monitored with the chemical actinometer pyridine/p-nitroanisole (PNA) (Laszakovits et al., 2017). Briefly, a solution containing 20 µM of recrystallized PNA and 0.25 mM of pyridine in nanopure water was irradiated for 5 h in the experimental conditions described above. Then, the PNA and pyridine were quantified via ultra-high-pressure liquid chromatography (Waters Acquity) equipped with a C18 column (BEH130 C18, 1.7 µm; 2.1 mm × 150 mm) and a photodiode array detector. The analyses were performed using a 40 : 60 A : B eluent mixture of high-performance liquid chromatography (HPLC)-grade solvent (A = 90 % (acetate buffer pH 6) + 10 % acetonitrile; B = 100 % acetonitrile), 5 µL injection volume and 0.2 mL min-1 flow rate. At these conditions, PNA was eluted at 4.2 min (detection at 310 nm) and pyridine at 2.7 min (detection at 250 nm).

The absolute spectral irradiance (Iλ) was obtained according to Iλ=s⋅Iλ,m, where Iλ,m (J nm-1 m-2 s-1) is the spectral irradiance measured in the photoreactor with a calibrated Jaz spectrophotometer (Ocean Optics) using 2 × 300 nm bulbs (Fig. S2). Iλ,m was also corrected for the absorption of the borosilicate glass occurring between 280 and 300 nm. The scaling factor s was calculated according to Eq. (1): 1s=kdeg,PNAPNA0lΦdeg,PNA∑λIλ,mfλ,PNAΔλ, where kdeg,PNA=(0.37±0.02) h-1 is the pseudo-first-order PNA degradation rate constant measured with 6×300 nm bulbs, [PNA]0=(17.4±0.4) µM is the starting PNA concentration, ΦPNA=0.076±0.008 is the direct photolysis quantum yield calculated according to Laszakovits et al. (2017), fλ,PNA is the absorptivity of the starting PNA solution, l is the path length (l=1 cm) and Δλ=1 nm. kdeg,PNA, [PNA]0 and ΦPNA are the average values measured for different days. fλ,PNA represents the fraction of light absorbed by the PNA starting solution, and it was calculated according to fλ,PNA=1-10-ελ,PNAPNA0l=1-10-Aλ,PNA, where ελ,PNA is the molar extinction coefficient of PNA and Aλ,PNA is the absorbance. This calculation provided an integrated irradiance (280–400 nm) of 64±4 J m-2 s-1 for 6×300 nm bulbs. The photoreactor-to-sunlight conversion factor, that is the irradiation time in the photoreactor equivalent to 1 h in natural sunlight, was obtained as the ratio of the rates of light absorption of a given DOM for the two different light sources (Rphotoreactorabs, Rsolarabs), according to Eq. (2). 2conversionfactor=RphotoreactorabsRsolarabs In order to get Rsolarabs, we simulated a solar spectrum (global horizontal irradiance, 300–400 nm, integrated irradiance of 59.2 J s-1 m-2) using SMARTS 2.9.5 (NREL) for a generic point at mid-latitudes (45∘) in summer (1 June, from 05:00 a.m. to 07:00 p.m.). Using the daily average value of 2.2 during daytime, 25 h of irradiation in the photoreactor with six UVB bulbs corresponds to 55 h of sunlight irradiation, which is equivalent to 4.6 d in the environment (assuming 12 daily hours of light on a clear day). The actinometry experiments are independent of temperature, as they depend on photon flux.

To study the effect of photochemistry on CCN and INPs, we sampled naturally occurring DOM from surface waters from the Great Dismal Swamp in Virginia, USA, in 2014 and in 2016 and at the Suwannee River, Florida, USA, in 2017. The natural DOM sampled from freshwater bodies is representative of a fraction of atmospheric organic aerosols directly relevant to lake spray aerosols and potentially prevalent in the aerosol phase over large freshwater bodies such as the Laurentian Great Lakes (Axson et al., 2016; Slade et al., 2010), the African Great Lakes and Lake Baikal. Furthermore, we used commercially available Suwannee River fulvic acid (SRFA), an established standard for complex organic matter and humic-like substances in both aquatic (Cory et al., 2009; Guerard et al., 2009) and atmospheric chemistry experiments (Chen and Valsaraj, 2007; Dinar et al., 2006; Frosch et al., 2011; Kessler et al., 2012; Ladino et al., 2016; Slade et al., 2017; Svenningsson et al., 2006; Wang and Knopf, 2011; Zelenay et al., 2011). To further bridge the gaps of aquatic and atmospheric chemistry, we used the same concentrations typically found in rivers and in cloud water, that is 20 mg L-1 (Cook et al., 2017; Johannesson et al., 2004). Organic aerosols containing humic-like substances may have lower average molecular weights and a lower aromatic moiety content than SRFA standards (Graber and Rudich, 2006), yet we complement our study with real surface-collected waters to circumvent IHSS extraction and processing of DOM. Ultimately, we argue that our DOM samples are naturally occurring in freshwater bodies and are adequate and alternative proxies for visualizing the photomineralization mechanism identified in this study.

Bulk DOM samples were irradiated for 25 h with UVB light in a photoreactor in order to simulate the photooxidation processes occurring in the atmosphere. Using a combination of chemical actinometry and radiometry measurements, we estimate that our 25 h exposure experiments in the photoreactor are equivalent to approximately 55 h of sunlight irradiation, which is further equivalent to 4.6 d in the environment. However, all experiments were conducted at 30–32 ∘C, and an extrapolation to colder and more relevant atmospheric temperatures would slow the rates of reactions of photochemical processes but would not affect the photon flux. Many photochemical processes have weak temperature dependences, and for instance hydrogen peroxide quantum yields are affected by a factor of 1.8 per 10 ∘C (Kieber et al., 2014). Thus, the exposure time corrected for lower temperature is relevant in the context of an approximately 1-week aerosol lifetime in the atmosphere.

The bulk solutions of irradiated DOM were subsequently analysed for CCN and INP activity. In all of our experiments, we employed concentrations between 14 and 20 mg of carbon per litre (mg C L-1), equivalent to 1160 to 1670 µM of organic carbon, which represents typical aquatic DOM conditions (Johannesson et al., 2004) as well as typical concentrations found in cloud water (Cook et al., 2017). An important caveat to highlight here is that our photochemical experiments were conducted in bulk. It is thus likely that the effect quantified does not necessarily equate to identical processes in the droplet and/or in the organic aerosol phase. However, we expect the photoproducts and mechanisms within the bulk to be occurring within an aerosol droplet, potentially at faster rates because of higher organic concentrations in the aqueous aerosol versus in the bulk. In addition, organic aerosols can also exist under different mixing states, phases and morphologies likely leading to matrix effects which were not captured in this work (Lignell et al., 2014; McDow et al., 1996).

To assess the ability of DOM to act as CCN, the κ values of each irradiated DOM sample were calculated, where higher κ values indicate a more effective CCN. During irradiation, we observed a clear increase in κ values, up to 2.5 times higher than the non-irradiated samples (Fig. 1). The change in CCN activity with increased UVB exposure was systematically observed across all of the naturally occurring samples tested, including the SRFA standard. In this study, κ values for photochemically exposed DOM ranged between 0.12 and 0.45 with an instrumental error of 0.01 and an experimental uncertainty of less than 15 %. These values compare to global ambient organic aerosol which has κ values between 0.15 and 0.30 (Schmale et al., 2017). We further note the seemingly large κ values for organic matter, specifically for the irradiated field-collected samples DS 2014, DS 2016 and SR 2017 (Fig. 1). We suggest that as the organic matter is photomineralized, the organic-to-inorganic ratio decreases, increasing κ values. Organic acids are also being formed and can contribute to increased κ values (see further discussion in Sect. 3.4.2). According to Mikhailov et al. (2013), κ is inversely proportional to molar mass, and, at equal particle density and at unity van't Hoff factor, a decrease in molar mass would lead to an increase in κ.

In particular, SRFA's CCN activity has been reported to have κ values of 0.06–0.08 (Dinar et al., 2006; Slade et al., 2017). In our experimental setup, SRFA has a κ of 0.12±0.02, and we attribute part of this discrepancy to the lower concentrations and the complete solubility of 20 mg C L-1 of SRFA used in our experiments. Concentrations above 5000 mg C L-1 were previously used, and such concentrations form a suspension rather than a homogeneously dissolved solution potentially leading to lower κ (Dinar et al., 2006; Slade et al., 2017). This possible concentration effect on CCN activity could arise from the interfacial molecular arrangement of the organic polymeric material (Ruehl et al., 2016), which leads to changes in surface tension and thus to κ values (Nozière et al., 2014; Ovadnevaite et al., 2017).

To probe the impact of photochemistry on ice nucleation, the field-collected DOM and SRFA were measured in our home-built DRoplet Ice Nuclei Counter Zurich (DRINCZ) (David et al., 2019). We report median freezing temperature (T50) values which correspond to the temperature at which 50 % of the wells filled with a DOM solution froze (Fig. 2a). The natural DOM samples from different locations and different years show frozen fractions (FFs) significantly above the background ultra-pure water, demonstrating ice nucleation activity (Fig. 2a). Specifically, DS 2014, DS 2016, SR 2017 and SRFA showed T50 values of -12.4±0.1, -7.8±0.3, -9.4±0.1 and -13.0±0.3 ∘C, respectively, which is well above the T50 value of ultra-pure water in DRINCZ (-22.5 ∘C). The steep curve of the FF curves exhibited by the samples suggests that the nature of the ice-nucleating material or active site is relatively homogeneous. These freezing temperatures are consistent with recently identified river nanoscale INPs (Knackstedt et al., 2018; Moffett et al., 2018). The FFs were then normalized by the organic carbon concentration in the solution measured by a TOC analyser to yield nm values as a function of temperature (Fig. S1). These nm values physically describe the ice nucleation active mass in units of number per milligram of carbon. We compare our nm values with those of the Wilson 2015 parameterization which represents the ice-nucleating ability in the immersion freezing mode of primarily organic material in the sea surface microlayer (Wilson et al., 2015) (Figs. 2c and S1). This comparison suggests that our DOM samples, which are smaller than 0.20 µm, could be accounting for a subset of the IN activity previously observed in the sea surface microlayer (Irish et al., 2019; McCluskey et al., 2018; Wilson et al., 2015). Furthermore, sea spray aerosols have recently been shown to be effective INPs when enriched in organics (DeMott et al., 2016; Ladino et al., 2016; Si et al., 2018; Wilson et al., 2015), consistent with our results. Soil organic matter, also containing humic-like substances, is also able to act as INPs in the immersion freezing mode (Hill et al., 2016; O'Sullivan et al., 2014). The SRFA was observed to have lower IN activity than the collected DOM samples, despite SR 2017 having been collected near the IHSS collection point at Suwannee River. This result suggests that extraction and processing of Suwannee River DOM to yield its fulvic acid component reduces its IN activity, potentially related to differences in organic carbon content (Botero et al., 2018). In other words, the fulvic acid is not as good as an INP compared to other components within DOM. Our results for SRFA are consistent with heterogeneous ice nucleation studies which used aerosolized SRFA in deposition freezing mode and found the organic material to be active (Wang and Knopf, 2011). Ultimately, DOM has the potential to nucleate ice at low to moderate (-4 to -12 ∘C) droplet supercooling through immersion freezing.

The DOM samples were then subjected to photochemical exposure in an identical setup to the CCN experiments, and the irradiated samples were measured on DRINCZ. Photochemistry decreased T50 values as a function of UVB irradiation, with an important impact on corresponding nm values (Fig. 2). Indeed, the 2 to 3 ∘C change in the T50 temperature observed in Fig. 2b leads to corresponding changes of up to 2 orders of magnitude in nm values between the 0 and the 25 h time points in Fig. 2c. In particular, for an nm of 100 ice nucleation sites per milligram of carbon, DS 2016 experienced a suppression in IN activity of up to 4 ∘C (Fig. 2c). The irradiated DOM samples led to different spreads of nm values, indicating DOM's composition-specific photochemistry response on immersion freezing, and warrants further study. Furthermore, the rate loss of IN activity measured as T50 could be quantified as a function of hours of UVB exposure. The slopes for photochemically exposed DS 2014, DS 2016, SR 2017 and SRFA were -0.054, -0.145, -0.066 and -0.100 ∘CT50 h-1, respectively (Fig. 2b). The average decrease in T50 values for irradiated DOM samples is thus -0.09 ∘CT50 h-1 of UVB light and equates to -0.04 ∘CT50 h-1 of sunlight at ground level in mid-latitudes (see methods for conversion of photochemical reactor exposure to sunlight exposure). These values could be useful to the atmospheric modelling community to investigate a lifetime-dependent IN activity of organic matter such as DOM or sea spray aerosols.

Although our study is the first to quantify the effect of photomineralization on the ice-nucleating ability of organic matter, others have identified specific organic molecules and components that could help explain the IN ability of DOM. Wilson et al. parameterized the ice-nucleating ability of sea surface microlayer samples using TOC and temperature (Wilson et al., 2015). However, this parameterization leads to higher estimates of nm than obtained for our DOM samples (Fig. 2c). This result suggests that nm does not efficiently constrain the IN activity of DOM. McCluskey et al. (2018) arrived at a similar conclusion for the IN activity of sea spray aerosol. The INPs measured in this study were soluble with an operational definition of passing through a 0.2 µm filter and could account for a subset of INPs identified in marine samples. In addition, soluble INPs have been shown to contain extracts from plant-based material, including cellulose (Hiranuma et al., 2015) and proteinaceous material (Augustin et al., 2013; Dreischmeier et al., 2017; Koop and Zobrist, 2009; O'Sullivan et al., 2015; Pummer et al., 2012, 2015; Wilson et al., 2015), which we know are present within our DOM samples. Finally, Gute and Abbatt observed that OH radical oxidation of pollen in deposition freezing required higher supersaturation ratio with respect to ice (Gute and Abbatt, 2018). Our results nicely corroborate with this observation and further suggest that not only OH radicals but photochemical processes in general are able to decrease the ice nucleation ability of organics, including pollen, in more than one freezing mode.

To understand the molecular origin of the observed photochemical impacts of DOM on cloud droplet (Fig. 1) and ice nucleation (Fig. 2), we tracked total carbon, absorbance, conductivity, pH, organic acids, CO and CO2 during irradiation.